#
✕ nick:bloom
#asahi-gpu · 2021-04-17 02:08
bloombut that might not be the allocation of threads
#asahi-gpu · 2021-04-17 02:08
bloomGranularity of 4 is almost certainly the allocation of registers
#asahi-gpu · 2021-04-17 02:08
bloomPossibly, possibly not.
#asahi-gpu · 2021-04-17 02:01
bloom"the shader cores... feature a constant execution and prefetch"
#asahi-gpu · 2021-04-17 01:58
bloomhttps://metalkit.org/2020/07/03/wwdc20-whats-new-in-metal.html has some goodies.
#asahi-gpu · 2021-04-17 01:57
bloom"Advanced Metal Shader Optimization" from WWDC'16 is still relevant :)
#asahi-gpu · 2021-04-17 01:55
bloomOh, even better maxTotalThreadsPerThreadgroup is literally in the Metal API. nice!
#asahi-gpu · 2021-04-17 01:48
bloomSo it ought to be able to figure out from the other direction like you've done with the cpu?
#asahi-gpu · 2021-04-17 01:48
bloomBut! Apple docs talk about occupancy perf counters?
#asahi-gpu · 2021-04-17 01:48
bloomThe cmdstream just specifies register count/4, but that seems more fine grained than real hw would implement
#asahi-gpu · 2021-04-17 01:47
bloomI'd like to know what the thresholds are for register usage --> thread occupancy
#asahi-gpu · 2021-04-17 01:47
bloomdougall: hey, I have a reversing question for you
#asahi-gpu · 2021-04-17 01:42
bloompiles of ALU
#asahi-gpu · 2021-04-16 22:46
bloomI would expect if you have fast imprecise transcendentals available, you'd use them and feed the estimate into Newton's method or something
#asahi-gpu · 2021-04-16 22:46
bloomwhat's more interesting to me is that they don't use the fast sine anywhere in it
#asahi-gpu · 2021-04-16 22:45
bloomthat it is..
#asahi-gpu · 2021-04-16 21:27
bloomdefault in Metal is fast::sin(x) which is super cheap
#asahi-gpu · 2021-04-16 21:19
bloom(~90 instructions)
#asahi-gpu · 2021-04-16 21:18
bloomThe assembly prduced for precise::sin(x) is horrifying
#asahi-gpu · 2021-04-16 20:13
bloom(was trying to find out if there's an xchg/swap instruction)
#asahi-gpu · 2021-04-16 20:12
bloomDunno if that's deliberate.
#asahi-gpu · 2021-04-16 20:12
bloomnote: Metal compiler likes bitop_mov more than xor
#asahi-gpu · 2021-04-16 19:57
bloomalso, modifiers are unsupported (both float and int mods)
#asahi-gpu · 2021-04-16 19:57
bloomto change sizes, iadd/fadd are used.
#asahi-gpu · 2021-04-16 19:57
bloomnote to self: `convert` is only for data type converts
#asahi-gpu · 2021-04-16 19:39
bloomInteresting... Metal implements unary fneg/fabs with bitop (twiddling the sign bit) instead of fadd.
#asahi-gpu · 2021-04-16 03:57
bloomwill test tomorrow 😴
#asahi-gpu · 2021-04-16 03:56
bloom" wrote a proper C encoder
#asahi-gpu · 2021-04-16 03:56
bloomdougall: "# okay, this is very lazy
#asahi-gpu · 2021-04-16 03:09
bloomso now all the known float ops are handled
#asahi-gpu · 2021-04-16 03:09
bloomand implemented in the compiler :)
#asahi-gpu · 2021-04-16 00:43
bloom:)
#asahi-gpu · 2021-04-15 21:50
bloomsimilarly interesting patterns with depth/stencil writeouts, et
#asahi-gpu · 2021-04-15 21:50
bloomand if you look at how MRT writeout works you can see a pair of bits moving along in those unk48 ("writeout" in dougall's)
#asahi-gpu · 2021-04-15 21:49
bloomI *suspect* that controls some detail of the tilebuffer
#asahi-gpu · 2021-04-15 21:49
bloomyep..
#asahi-gpu · 2021-04-15 18:21
bloom(Up to constants, sin_pt_2 is the "sinc" function.)
#asahi-gpu · 2021-04-15 18:18
bloomAnd that's how AGX computes sine and cosine!
#asahi-gpu · 2021-04-15 18:17
bloom(This requires the value to be finite, since Inf * 0 and NaN * 0 are both NaN under standard IEEE 754 rules.)
#asahi-gpu · 2021-04-15 18:16
bloom(Is sin_pt_2(0) undefined? Maybe, maybe not. It doesn't matter, as long as it's value satisfies sin_pt_2(0) * 0 = 0 since sin(0) = 0.)
#asahi-gpu · 2021-04-15 18:15
bloomsin_pt_2(x) = sin((pi/2) * |x|) / |x|
#asahi-gpu · 2021-04-15 18:15
bloomAs well as:
#asahi-gpu · 2021-04-15 18:15
bloom{ fract(x) - 1 if 3 <= x < 4
#asahi-gpu · 2021-04-15 18:15
bloom{ - fract(x) if 2 <= x < 3
#asahi-gpu · 2021-04-15 18:15
bloom{ 1 - fract(x) if 1 <= x < 2
#asahi-gpu · 2021-04-15 18:14
bloom{ fract(x) if 0 <= x < 1
#asahi-gpu · 2021-04-15 18:14
bloomsin_pt_1(x) =
#asahi-gpu · 2021-04-15 18:14
bloomOr more clearly:
#asahi-gpu · 2021-04-15 18:14
bloom{ x - 4 if 3 <= x < 4
#asahi-gpu · 2021-04-15 18:14
bloom{ 2 - x if 2 <= x < 3

plus ›